Breast cancer is a heterogeneous disease that can be stratified in at least 10 different subtypes. We present here a platform for derivation of preclinical models based on patient-derived tumor xenografts (PDTXs) that represent these subgroups. These models preserve the transcriptome, methylome, copy-number, and mutational landscape features of the tumor of origin through different passaging. Furthermore, the intratumoral composition of these models is formed by communities of clones very similar to the ones present in the originating tumor. Finally, we show that short-term cultures of cells from these models (PDTX-derived tumor cells, PDTCs) also preserve the molecular features of the tumor and can be used for high-throughput drug testing of single compounds or combinations, with high reproducibility and clinical predictive power.